🧪 LLM Evaluation Gate: Batch Size, Cache Hits and Queue Latency in 3D
A Blender-built evaluation line where test cases queue on a belt, hit a cache cabinet or wait for a GPU server that processes them in batches: change the batch cap, cache hit rate and shadow traffic and watch latency and GPU cost respond, with five independent numerical checks (M/D/1, Markov chain, utilisation identity, exact batch counts, binomial cache misses).
A Blender-built evaluation line where test cases queue on a belt, hit a cache cabinet or wait for a GPU server that processes them in batches: change the batch cap, cache hit rate and shadow traffic and watch latency and GPU cost respond, with five independent numerical checks (M/D/1, Markov chain, utilisation identity, exact batch counts, binomial cache misses).
3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install